The Federal Government's AI Surveillance Buildout Is Bigger Than Anyone Realized

The Federal Government's AI Surveillance Buildout Is Bigger Than Anyone Realized

Autonomous Eyes on the Border

Two weeks ago, U.S. Customs and Border Protection awarded General Dynamics Information Technology a $71 million task order for autonomous surveillance towers slated for deployment along the southern border. The contract is part of a broader indefinite-delivery, indefinite-quantity agreement worth up to $1.8 billion that CBP launched in 2023 to modernize its tower network. GDIT certified the towers under the One Big Beautiful Bill Act, which specifically requires autonomous systems to apply machine learning, computer vision, or related algorithms to detect, classify, and track targets in real time, without continuous human oversight.

The towers use edge AI, electro-optical sensors, radar, and LIDAR, supported by 5G and satellite communications. According to GDIT, they can track hundreds of targets simultaneously. The Department of Homeland Security planned to install AI upgrades in 148 existing camera towers this year and add 50 next-generation units. Three vendors competed for the work: GDIT, Anduril, and Elbit.

The supply chain is already a constraint. CBP Commissioner Rodney Scott told lawmakers in April that the challenge was less about signing contracts and more about whether vendors could deliver. GDIT's vice president for border security confirmed that chip shortages tied to the AI boom have created real logistics problems in the past six months. The infrastructure is being ordered faster than it can be built.

The Facial Recognition Stack

In February 2026, CBP signed a $225,000 contract with Clearview AI, giving intelligence analysts at the agency's National Targeting Center access to a database of more than 60 billion images scraped from social media platforms, websites, and public applications. The contract is structured around "tactical targeting," a term CBP uses for real-time identification of individuals during enforcement operations.

This was not CBP's first purchase from Clearview. The company had already run a pilot with the agency in 2025. ICE signed its own Clearview contract for $9.2 million in September 2025, and added a $3.75 million contract described as ICE's largest Clearview purchase to date. The contracts sit alongside a broader surveillance stack that includes Palantir's ImmigrationOS, Babel Street for social media monitoring, and Penlink for tracking mobile devices. Senators Mark Warner and Tim Kaine wrote to DHS demanding an investigation into the agency's use of these tools in January 2026.

CBP's stated guardrail on Clearview is that no enforcement action can be taken based solely on leads generated by the system, and that all identifications require further investigation. The agency said the technology was tuned during its pilot phase to limit misidentification. What it has not disclosed is the error rate of the tuned system or how often its outputs are used as the primary signal prompting a field stop.

ICE has gone further. Internal footage obtained by Media 404 showed ICE officers using a facial recognition app called Mobile Fortify to check the citizenship status of teenagers who were not carrying identification. The app draws from more than 200 million images stored across DHS, FBI, and State Department databases, and appears to connect to data aggregation tools that pull across multiple government systems. That use case does not appear in any official AI inventory as an active enforcement deployment.

ImmigrationOS and the Data Integration Problem

In April 2025, Palantir received a $30 million contract to build ImmigrationOS for ICE. The platform promises what Palantir describes as "granular tracking" of immigrants through every stage of the immigration lifecycle, including real-time monitoring of self-deportations. The system connects to ICE's Investigative Case Management platform, which has access to data from across the federal government.

The Trump administration's Department of Government Efficiency has been working in parallel to centralize data from agencies that have historically maintained separate records. According to reporting from Wired cited by Brookings, DOGE sought to integrate data from the Social Security Administration and the IRS into USCIS's "data lake," which contains information on immigration cases. Separate reports indicated that ICE and DOGE requested access to Medicare data to obtain addresses of immigrants, despite non-residents being ineligible for the program.

The practical result of these integrations is that a system originally designed for immigration case management now draws on tax records, Social Security files, and health program data. The American Immigration Council noted in a 2025 review that DHS had marked several older AI programs as "inactive" in its use case inventory, but that the underlying capabilities of those programs had been consolidated into the large vendor platforms rather than discontinued. The Clearview, Palantir, and Paragon contracts effectively absorbed the functions of programs that no longer appear in official reporting.

Inside the Agencies: Internal AI Tools

The AI buildout is not only facing outward. CBP has deployed an internal chatbot called chatCBP, built on a large language model, for tasks including document summarization and multi-file analysis. CBP's chief technology officer publicly described the tool at a conference in May 2025. The agency declined to disclose which underlying model it uses. Other agencies have been more transparent: the General Services Administration offers staff access to models from Meta and Anthropic, and the State Department's StateChat runs on Palantir and Azure OpenAI.

CBP is also using Google's Vertex AI to allow staff to search across disparate internal data sources and pull them into a unified view. CBP's deputy assistant commissioner called the capability an "absolute game-changer," particularly for agents in remote locations who previously had limited access to centralized data. The agency has been running AI-powered video analytics at ports of entry since at least 2023, using machine learning models to flag anomalies in streaming footage and alert operators to potential threats.

The U.S. Citizenship and Immigration Services has been running a system called the Evidence Classifier, which uses machine learning to automatically tag individual pages of immigration case files with evidence categories, reducing the manual work involved in adjudicating high-volume case types. The agency has also explored using AI to detect inconsistencies across filings, including changes in job duties, salary information, or profile details between applications.

At the IRS, AI adoption is moving more slowly. The agency lost 27% of its workforce last year through voluntary separation and early retirement, and has been reassigning IT and HR staff to taxpayer services on 120-day rotations to compensate. Its current AI initiative focuses on accelerating training for new customer service representatives, a process that currently takes 14 weeks. That work remains at the proof-of-concept stage.

What the Spending Numbers Show

A May 2026 Brookings Institution analysis of federal AI contracts offers a useful baseline. From 2022 to 2024, the value of federally obligated AI contracts grew from $261 million to $675 million, a 150% increase over two years. Between 2024 and 2026, obligated spending jumped to $7.2 billion, a 966% increase. Potential contract award values reached $91.8 billion.

The Department of Defense accounts for 98.9% of that potential value, with $90.7 billion in AI contract commitments. The scale of defense spending is large enough that DHS, HHS, NASA, and every other civilian agency collectively represent less than 1.1% of the federal AI market by potential contract value. HHS saw a 448% increase, from $27 million in 2024 to $138 million in 2026, but those figures barely register against the defense numbers.

The number of federal agencies with AI contracts grew from 17 in 2022 to 28 in 2026, with new entrants including the Executive Office of the President. A GAO report published in April 2026 found that agencies consistently struggle with AI procurement: they have difficulty finding data scientists to evaluate vendor proposals, they cannot reliably assess AI-related costs, and they do not systematically collect or share lessons learned from past purchases. The GAO recommended that agencies develop common practices for documenting and distributing procurement experience before committing to further expansion.

The Governance Gap

The oversight machinery has not kept pace with the deployment rate. Senator Ed Markey has introduced legislation to bar ICE and CBP from using facial recognition technology entirely. Bicameral legislation introduced by Senators Markey, Jeff Merkley, and Ron Wyden calls for a full ban on the technology at DHS enforcement agencies.

The OMB published two AI governance memos in April 2025 covering acquisition and risk management, and a third in December 2025 addressing "unbiased AI principles" for large language models. President Trump signed a new executive order on June 2, 2026, establishing cybersecurity mandates for frontier AI deployment and a voluntary framework for federal AI adoption. The June order reinforces the administration's position that regulatory friction should not slow AI deployment in government.

What the policy framework does not address clearly is the question of when a warrant is required for biometric scanning of citizens or non-citizens, a gap that courts across different jurisdictions have resolved inconsistently. CBP operates under legal authorities that permit warrantless stops and searches within 100 miles of any U.S. land border, a zone that encompasses most of the country's largest cities. The combination of those legacy authorities and new AI-powered surveillance tools produces a legal environment that independent auditors have not yet examined in any systematic way.

What Comes Next

The trajectory across these agencies follows a recognizable pattern. Technologies that entered government use as narrow pilots, facial recognition for specific investigative leads, algorithmic scoring for case prioritization, automated cargo scanning for high-volume ports, have moved into operational infrastructure at a speed that procurement paperwork and oversight frameworks were not designed to handle. The GAO's April finding that agencies are not documenting or sharing what they learn from AI acquisitions suggests that the institutional knowledge needed to govern these systems is not accumulating as fast as the systems themselves are being deployed.

The fundamental question for enterprises watching this space is whether the government's current approach, scaling commercial AI tools into enforcement and administrative workflows with limited independent evaluation, produces the accountability that high-stakes decision-making requires. The towers going up on the southern border, the facial recognition contracts now moving from pilot to operational, and the data integrations connecting previously siloed agency records represent a significant change in the infrastructure of federal enforcement. What they do not yet represent is a clear public accounting of how well these systems perform when the subject of the decision is a person.